The default len of request uri in log messages is 1024. In some use
cases, you need to keep the long trail of GET parameters. The only
way to increase this len is to recompile with DEFINE=-DREQURI_LEN=2048.
This commit introduces a tune.http.logurilen configuration directive,
allowing to tune this at runtime.
This option exits every workers when one of the current workers die.
It allows you to monitor the master process in order to relaunch
everything on a failure.
For example it can be used with systemd and Restart=on-failure in a spec
file.
This commit remove the -Ds systemd mode in HAProxy in order to replace
it by a more generic master worker system. It aims to replace entirely
the systemd wrapper in the near future.
The master worker mode implements a new way of managing HAProxy
processes. The master is in charge of parsing the configuration
file and is responsible for spawning child processes.
The master worker mode can be invoked by using the -W flag. It can be
used either in background mode (-D) or foreground mode. When used in
background mode, the master will fork to daemonize.
In master worker background mode, chroot, setuid and setgid are done in
each child rather than in the master process, because the master process
will still need access to filesystem to reload the configuration.
This patch changes the stats socket rights for allowing the sending of
listening sockets.
The previous behavior was to allow any unix stats socket with admin
level to send sockets. It's not possible anymore, you have to set this
option to activate the socket sending.
Example:
stats socket /var/run/haproxy4.sock mode 666 expose-fd listeners level user process 4
The current level variable use only 2 bits for storing the 3 access
level (user, oper and admin).
This patch add a bitmask which allows to use the remaining bits for
other usage.
When running with multiple process, if some proxies are just assigned
to some processes, the other processes will just close the file descriptors
for the listening sockets. However, we may still have to provide those
sockets when reloading, so instead we just try hard to pretend those proxies
are dead, while keeping the sockets opened.
A new global option, no-reused-socket", has been added, to restore the old
behavior of closing the sockets not bound to this process.
When SIGUSR1 is received, haproxy enters in soft-stop and quits when no
connection remains.
It can happen that the instance remains alive for a long time, depending
on timeouts and traffic. This option ensures that soft-stop won't run
for too long.
Example:
global
hard-stop-after 30s # Once in soft-stop, the instance will remain
# alive for at most 30 seconds.
Historically a lot of SSL global settings were stored into the global
struct, but we've reached a point where there are 3 ifdefs in it just
for this, and others in haproxy.c to initialize it.
This patch moves all the private fields to a new struct "global_ssl"
stored in ssl_sock.c. This includes :
char *crt_base;
char *ca_base;
char *listen_default_ciphers;
char *connect_default_ciphers;
int listen_default_ssloptions;
int connect_default_ssloptions;
int tune.sslprivatecache; /* Force to use a private session cache even if nbproc > 1 */
unsigned int tune.ssllifetime; /* SSL session lifetime in seconds */
unsigned int tune.ssl_max_record; /* SSL max record size */
unsigned int tune.ssl_default_dh_param; /* SSL maximum DH parameter size */
int tune.ssl_ctx_cache; /* max number of entries in the ssl_ctx cache. */
The "tune" part was removed (useless here) and the occasional "ssl"
prefixes were removed as well. Thus for example instead of
global.tune.ssl_default_dh_param
we now have :
global_ssl.default_dh_param
A few initializers were present in the constructor, they could be brought
back to the structure declaration.
A few other entries had to stay in global for now. They concern memory
calculationn (used in haproxy.c) and stats (used in stats.c).
The code is already much cleaner now, especially for global.h and haproxy.c
which become readable.
This finishes to clean up the zlib-specific parts. It also unbreaks recent
commit b97c6fb ("CLEANUP: compression: use the build options list to report
the algos") which broke USE_ZLIB due to MAXWBITS not being defined anymore
in haproxy.c.
We replaced global.deviceatlas with global_deviceatlas since there's no need
to store all this into the global section. This removes the last #ifdefs,
and now the code is 100% self-contained in da.c. The file da.h was now
removed because it was only used to load dac.h, which is more easily
loaded directly from da.c. It provides another good example of how to
integrate code in the future without touching the core parts.
We replaced global._51degrees with global_51degrees since there's no need
to store all this into the global section. This removes the last #ifdefs,
and now the code is 100% self-contained in 51d.c. The file 51d.h was now
removed because it was only used to load 51Degrees.h, which is more easily
loaded from 51d.c. It provides a good example of how to integrate code in
the future without touching the core parts.
We replaced global.wurfl with global_wurfl since there's no need to store
all this into the global section. This removes the last #ifdefs, and now
the code is 100% self-contained in wurfl.c. It provides a good example of
how to integrate code in the future without touching the core parts.
The 3 device detection engines stop at the same place in deinit()
with the usual #ifdefs. Similar to the other functions we can have
some late deinitialization functions. These functions do not return
anything however so we have to use a different type.
There's a significant amount of late initialization calls which are
performed after the point where we exit in check mode. These calls
are used to allocate resource and perform certain slow operations.
Let's have a way to register some functions which need to be called
there instead of having this multitude of #ifdef in the init path.
Many extensions now report some build options to ease debugging, but
this is now being done at the expense of code maintainability. Let's
provide a registration function to do this so that we can start to
remove most of the #ifdefs from haproxy.c (18 currently just for a
single function).
Now it is possible to use variables attached to a process. The scope name is
'proc'. These variables are released only when HAProxy is stopped.
'tune.vars.proc-max-size' directive has been added to confiure the maximum
amount of memory used by "proc" variables. And because memory accounting is
hierachical for variables, memory for "proc" vars includes memory for "sess"
vars.
It is very common when validating a configuration out of production not to
have access to the same resolvers and to fail on server address resolution,
making it difficult to test a configuration. This option simply appends the
"none" method to the list of address resolution methods for all servers,
ensuring that even if the libc fails to resolve an address, the startup
sequence is not interrupted.
The only reason wurfl/wurfl.h was needed outside of wurfl.c was to expose
wurfl_handle which is a pointer to a structure, referenced by global.h.
By just storing a void* there instead, we can confine all wurfl code to
wurfl.c, which is really nice.
WURFL is a high-performance and low-memory footprint mobile device
detection software component that can quickly and accurately detect
over 500 capabilities of visiting devices. It can differentiate between
portable mobile devices, desktop devices, SmartTVs and any other types
of devices on which a web browser can be installed.
In order to add WURFL device detection support, you would need to
download Scientiamobile InFuze C API and install it on your system.
Refer to www.scientiamobile.com to obtain a valid InFuze license.
Any useful information on how to configure HAProxy working with WURFL
may be found in:
doc/WURFL-device-detection.txt
doc/configuration.txt
examples/wurfl-example.cfg
Please find more information about WURFL device detection API detection
at https://docs.scientiamobile.com/documentation/infuze/infuze-c-api-user-guide
With Linux officially introducing SO_REUSEPORT support in 3.9 and
its mainstream adoption we have seen more people running into strange
SO_REUSEPORT related issues (a process management issue turning into
hard to diagnose problems because the kernel load-balances between the
new and an obsolete haproxy instance).
Also some people simply want the guarantee that the bind fails when
the old process is still bound.
This change makes SO_REUSEPORT configurable, introducing the command
line argument "-dR" and the noreuseport configuration directive.
A backport to 1.6 should be considered.
Trie now uses a dataset structure just like Pattern, so this has been
defined in includes/types/global.h for both Pattern and Trie where it
was just Pattern.
In src/51d.c all functions used by the Trie implementation which need a
dataset as an argument now use the global dataset. The
fiftyoneDegreesDestroy method has now been replaced with
fiftyoneDegreesDataSetFree which is common to Pattern and Trie. In
addition, two extra dataset init status' have been added to the switch
statement in init_51degrees.
When memmax is forced using "-m", the per-process memory limit is enforced
using setrlimit(), but this value is not used to compute the automatic
maxconn limit. In addition, the per-process memory limit didn't consider
the fact that the shared SSL cache only needs to be accounted once.
The doc was also fixed to clearly state that "-m" is global and not per
process. It makes sense because people who use -m want to protect the
system's resources regardless of whatever appears in the configuration.
Michael Ezzell reported a bug causing haproxy to segfault during startup
when trying to send syslog message from Lua. The function __send_log() can
be called with *p that is NULL and/or when the configuration is not fully
parsed, as is the case with Lua.
This patch fixes this problem by using individual vectors instead of the
pre-generated strings log_htp and log_htp_rfc5424.
Also, this patch fixes a problem causing haproxy to write the wrong pid in
the logs -- the log_htp(_rfc5424) strings were generated at the haproxy
start, but "pid" value would be changed after haproxy is started in
daemon/systemd mode.
Added support for version 3.2 of 51Degrees C library.
Added fields to store HTTP header names important to device detection
other than User-Agent.
Included a pool of worksets for use with Pattern device detection.
This new global section directive is used to store the path to the file
where HAProxy will be able to retrieve server states across reloads.
The file pointed by this path is used to store a file which can contains
state of all servers from all backends.
This new global directive can be used to provide a base directory where
all the server state files could be loaded.
If a server state file name starts with a slash '/', then this directive
must not be applied.
This was the first transparent proxy technology supported by haproxy
circa 2005 but it was obsoleted in 2007 by Tproxy 4.0 which removed a
lot of the earlier versions' shortcomings and was finally merged into
the kernel. Since nobody has been using cttproxy for many years now
and nobody has even just tried to compile the files, it's time to
remove it. The doc was updated as well.
This cache is used by 51d converter. The input User-Agent string, the
converter args and a random seed are used as a hashing key. The cached
entries contains a pointer to the resulting string for specific
User-Agent string detection.
The cache size can be tuned using 51degrees-cache-size parameter.
Moved 51Degrees code from src/haproxy.c, src/sample.c and src/cfgparse.c
into a separate files src/51d.c and include/import/51d.h.
Added two new functions init_51degrees() and deinit_51degrees(), updated
Makefile and other code reorganizations related to 51Degrees.
Implementation of a DNS client in HAProxy to perform name resolution to
IP addresses.
It relies on the freshly created UDP client to perform the DNS
resolution. For now, all UDP socket calls are performed in the
DNS layer, but this might change later when the protocols are
extended to be more suited to datagram mode.
A new section called 'resolvers' is introduced thanks to this patch. It
is used to describe DNS servers IP address and also many parameters.
With this patch, it is possible to configure HAProxy to forge the SSL
certificate sent to a client using the SNI servername. We do it in the SNI
callback.
To enable this feature, you must pass following BIND options:
* ca-sign-file <FILE> : This is the PEM file containing the CA certitifacte and
the CA private key to create and sign server's certificates.
* (optionally) ca-sign-pass <PASS>: This is the CA private key passphrase, if
any.
* generate-certificates: Enable the dynamic generation of certificates for a
listener.
Because generating certificates is expensive, there is a LRU cache to store
them. Its size can be customized by setting the global parameter
'tune.ssl.ssl-ctx-cache-size'.
This diff is the raw C struct definition of all DeviceAtlas module
data needed added to the main global struct haproxy configuration.
The three first members are needed for both init and deinit phases
as some dynamic memory allocations are done.
The useragentid serves to hold during the whole lifecycle of the
module the User-Agent HTTP Header identifier from the DeviceAtlas
data during the init process.
These ones were already obsoleted in 1.4, marked for removal in 1.5,
and not documented anymore. They used to emit warnings, and do still
require quite some code to stay in place. Let's remove them now.
The principle of this cache is to have a global cache for all pattern
matching operations which rely on lists (reg, sub, dir, dom, ...). The
input data, the expression and a random seed are used as a hashing key.
The cached entries contains a pointer to the expression and a revision
number for that expression so that we don't accidently used obsolete
data after a pattern update or a very unlikely hash collision.
Regarding the risk of collisions, 10k entries at 10k req/s mean 1% risk
of a collision after 60 years, that's already much less than the memory's
reliability in most machines and more durable than most admin's life
expectancy. A collision will result in a valid result to be returned
for a different entry from the same list. If this is not acceptable,
the cache can be disabled using tune.pattern.cache-size.
A test on a file containing 10k small regex showed that the regex
matching was limited to 6k/s instead of 70k with regular strings.
When enabling the LRU cache, the performance was back to 70k/s.
An SSL connection takes some memory when it exists and during handshakes.
We measured up to 16kB for an established endpoint, and up to 76 extra kB
during a handshake. The SSL layer stores these values into the global
struct during initialization. If other SSL libs are used, it's easy to
change these values. Anyway they'll only be used as gross estimates in
order to guess the max number of SSL conns that can be established when
memory is constrained and the limit is not set.
We'll need to know the number of SSL connections, their use and their
cost soon. In order to avoid getting tons of ifdefs everywhere, always
export SSL information in the global section. We add two flags to know
whether or not SSL is used in a frontend and in a backend.
This setting is used to limit memory usage without causing the alloc
failures caused by "-m". Unexpectedly, tests have shown a performance
boost of up to about 18% on HTTP traffic when limiting the number of
buffers to about 10% of the amount of concurrent connections.
tune.buffers.limit <number>
Sets a hard limit on the number of buffers which may be allocated per process.
The default value is zero which means unlimited. The minimum non-zero value
will always be greater than "tune.buffers.reserve" and should ideally always
be about twice as large. Forcing this value can be particularly useful to
limit the amount of memory a process may take, while retaining a sane
behaviour. When this limit is reached, sessions which need a buffer wait for
another one to be released by another session. Since buffers are dynamically
allocated and released, the waiting time is very short and not perceptible
provided that limits remain reasonable. In fact sometimes reducing the limit
may even increase performance by increasing the CPU cache's efficiency. Tests
have shown good results on average HTTP traffic with a limit to 1/10 of the
expected global maxconn setting, which also significantly reduces memory
usage. The memory savings come from the fact that a number of connections
will not allocate 2*tune.bufsize. It is best not to touch this value unless
advised to do so by an haproxy core developer.
We've already experimented with three wake up algorithms when releasing
buffers : the first naive one used to wake up far too many sessions,
causing many of them not to get any buffer. The second approach which
was still in use prior to this patch consisted in waking up either 1
or 2 sessions depending on the number of FDs we had released. And this
was still inaccurate. The third one tried to cover the accuracy issues
of the second and took into consideration the number of FDs the sessions
would be willing to use, but most of the time we ended up waking up too
many of them for nothing, or deadlocking by lack of buffers.
This patch completely removes the need to allocate two buffers at once.
Instead it splits allocations into critical and non-critical ones and
implements a reserve in the pool for this. The deadlock situation happens
when all buffers are be allocated for requests pending in a maxconn-limited
server queue, because then there's no more way to allocate buffers for
responses, and these responses are critical to release the servers's
connection in order to release the pending requests. In fact maxconn on
a server creates a dependence between sessions and particularly between
oldest session's responses and latest session's requests. Thus, it is
mandatory to get a free buffer for a response in order to release a
server connection which will permit to release a request buffer.
Since we definitely have non-symmetrical buffers, we need to implement
this logic in the buffer allocation mechanism. What this commit does is
implement a reserve of buffers which can only be allocated for responses
and that will never be allocated for requests. This is made possible by
the requester indicating how much margin it wants to leave after the
allocation succeeds. Thus it is a cooperative allocation mechanism : the
requester (process_session() in general) prefers not to get a buffer in
order to respect other's need for response buffers. The session management
code always knows if a buffer will be used for requests or responses, so
that is not difficult :
- either there's an applet on the initiator side and we really need
the request buffer (since currently the applet is called in the
context of the session)
- or we have a connection and we really need the response buffer (in
order to support building and sending an error message back)
This reserve ensures that we don't take all allocatable buffers for
requests waiting in a queue. The downside is that all the extra buffers
are really allocated to ensure they can be allocated. But with small
values it is not an issue.
With this change, we don't observe any more deadlocks even when running
with maxconn 1 on a server under severely constrained memory conditions.
The code becomes a bit tricky, it relies on the scheduler's run queue to
estimate how many sessions are already expected to run so that it doesn't
wake up everyone with too few resources. A better solution would probably
consist in having two queues, one for urgent requests and one for normal
requests. A failed allocation for a session dealing with an error, a
connection event, or the need for a response (or request when there's an
applet on the left) would go to the urgent request queue, while other
requests would go to the other queue. Urgent requests would be served
from 1 entry in the pool, while the regular ones would be served only
according to the reserve. Despite not yet having this, it works
remarkably well.
This mechanism is quite efficient, we don't perform too many wake up calls
anymore. For 1 million sessions elapsed during massive memory contention,
we observe about 4.5M calls to process_session() compared to 4.0M without
memory constraints. Previously we used to observe up to 16M calls, which
rougly means 12M failures.
During a test run under high memory constraints (limit enforced to 27 MB
instead of the 58 MB normally needed), performance used to drop by 53% prior
to this patch. Now with this patch instead it *increases* by about 1.5%.
The best effect of this change is that by limiting the memory usage to about
2/3 to 3/4 of what is needed by default, it's possible to increase performance
by up to about 18% mainly due to the fact that pools are reused more often
and remain hot in the CPU cache (observed on regular HTTP traffic with 20k
objects, buffers.limit = maxconn/10, buffers.reserve = limit/2).
Below is an example of scenario which used to cause a deadlock previously :
- connection is received
- two buffers are allocated in process_session() then released
- one is allocated when receiving an HTTP request
- the second buffer is allocated then released in process_session()
for request parsing then connection establishment.
- poll() says we can send, so the request buffer is sent and released
- process session gets notified that the connection is now established
and allocates two buffers then releases them
- all other sessions do the same till one cannot get the request buffer
without hitting the margin
- and now the server responds. stream_interface allocates the response
buffer and manages to get it since it's higher priority being for a
response.
- but process_session() cannot allocate the request buffer anymore
=> We could end up with all buffers used by responses so that none may
be allocated for a request in process_session().
When the applet processing leaves the session context, the test will have
to be changed so that we always allocate a response buffer regardless of
the left side (eg: H2->H1 gateway). A final improvement would consists in
being able to only retry the failed I/O operation without waking up a
task, but to date all experiments to achieve this have proven not to be
reliable enough.
Adds global statements 'ssl-default-server-options' and
'ssl-default-bind-options' to force on 'server' and 'bind' lines
some ssl options.
Currently available options are 'no-sslv3', 'no-tlsv10', 'no-tlsv11',
'no-tlsv12', 'force-sslv3', 'force-tlsv10', 'force-tlsv11',
'force-tlsv12', and 'no-tls-tickets'.
Example:
global
ssl-default-server-options no-sslv3
ssl-default-bind-options no-sslv3
With all the goodies supported by logformat, people find that the limit
of 1024 chars for log lines is too short. Some servers do not support
larger lines and can simply drop them, so changing the default value is
not always the best choice.
This patch takes a different approach. Log line length is specified per
log server on the "log" line, with a value between 80 and 65535. That
way it's possibly to satisfy all needs, even with some fat local servers
and small remote ones.
When no static DH parameters are specified, this patch makes haproxy
use standardized (rfc 2409 / rfc 3526) DH parameters with prime lenghts
of 1024, 2048, 4096 or 8192 bits for DHE key exchange. The size of the
temporary/ephemeral DH key is computed as the minimum of the RSA/DSA server
key size and the value of a new option named tune.ssl.default-dh-param.
One important aspect of SSL performance tuning is the cache size,
but there's no metric to know whether it's large enough or not. This
commit introduces two counters, one for the cache lookups and another
one for cache misses. These counters are reported on "show info" on
the stats socket. This way, it suffices to see the cache misses
counter constantly grow to know that a larger cache could possibly
help.
It's commonly needed to know how many SSL asymmetric keys are computed
per second on either side (frontend or backend), and to know the SSL
session reuse ratio. Now we compute these values and report them in
"show info".
Some consistency checks cannot be performed between frontends, backends
and peers at the moment because there is no way to check for intersection
between processes bound to some processes when the number of processes is
higher than the number of bits in a word.
So first, let's limit the number of processes to the machine's word size.
This means nbproc will be limited to 32 on 32-bit machines and 64 on 64-bit
machines. This is far more than enough considering that configs rarely go
above 16 processes due to scalability and management issues, so 32 or 64
should be fine.
This way we'll ensure we can always build a mask of all the processes a
section is bound to.
